Deciphering the principles of the RNA editing code via large-scale systematic probing

•A-to-I editing is strongly induced 30–35 bp away from structural disruptions•A-to-I symmetrically introduced on opposing sites of a dsRNA•Structural disruptions by give rise to the next event•Distant mismatches increase targeted efficiency Adenosine-to-inosine catalyzed ADAR1 at thousands transcriptome-wide. Despite intense interest in physiological, bioengineering, and therapeutic perspectives, rules substrate selection are poorly understood. Here, we used large-scale systematic probing ?2,000 synthetic constructs explore structure sequence context determining editability. We uncover two layers formation propagation A-to-I editing, independent sequence. First, robustly fixed intervals 35 upstream 30 downstream disruptions. Second, opposite double-stranded structure. Our findings suggest recursive model for RNA whereby alteration one site iteratively gives an additional periodicity, serving as basis along across both strands structures. Across human transcriptome, hundreds adenosines deaminated into inosines (Porath et al., 2014Porath H.T. Carmi S. Levanon E.Y. A genome-wide map hyper-edited reveals numerous new sites.Nat. Commun. 2014; 5: 4726Crossref PubMed Scopus (123) Google Scholar; Ramaswami Li, 2014Ramaswami G. Li J.B. RADAR: rigorously annotated database editing.Nucleic Acids Res. 42: D109-D113Crossref (316) Scholar). The vast majority these events ADAR1, which preferentially edits strong (dsRNA) structures non-coding elements, mainly inverted-repeat (IR) Alus (Bazak 2014aBazak L. Haviv A. Barak M. Jacob-Hirsch J. Deng P. Zhang R. Isaacs F.J. Rechavi Eisenberg E. occurs over hundred million genomic sites, located genes.Genome 24: 365-376Crossref (315) Scholar, Bazak 2014bBazak Genome-wide analysis Alu editability.Nucleic 6876-6884Crossref (68) 2004Levanon Yelin Nemzer Hallegger Shemesh Fligelman Z.Y. Shoshan Pollock S.R. Sztybel D. al.Systematic identification abundant transcriptome.Nat. Biotechnol. 2004; 22: 1001-1005Crossref (607) Morse 2002Morse D.P. Aruscavage P.J. Bass B.L. hairpins noncoding regions brain Caenorhabditis elegans mRNA edited adenosine deaminases that act RNA.Proc. Natl. Acad. Sci. USA. 2002; 99: 7906-7911Crossref (182) Nishikura, 2016Nishikura K. coding RNAs ADARs.Nat. Rev. Mol. Cell Biol. 2016; 17: 83-96Crossref (438) Several studies main role disrupt such long dsRNA, thereby preventing their recognition melanoma differentiation-associated protein 5 (MDA5) subsequent activation interferon pathway (Ahmad 2018Ahmad Mu X. Yang F. Greenwald Park J.W. Jacob C.-Z. Hur Breaching Self-Tolerance Duplex Underlies MDA5-Mediated Inflammation.Cell. 2018; 172: 797-810.e13Abstract Full Text PDF (171) Hartner 2009Hartner J.C. Walkley C.R. Lu Orkin S.H. essential maintenance hematopoiesis suppression signaling.Nat. Immunol. 2009; 10: 109-115Crossref (318) Liddicoat 2015Liddicoat B.J. Piskol Chalk A.M. Higuchi Seeburg P.H. prevents MDA5 sensing endogenous dsRNA nonself.Science. 2015; 349: 1115-1120Crossref (408) Pestal 2015Pestal Funk C.C. Snyder J.M. Price N.D. Treuting P.M. Stetson D.B. Isoforms RNA-Editing Enzyme Independently Control Nucleic Acid Sensor MDA5-Driven Autoimmunity Multi-organ Development.Immunity. 43: 933-944Abstract (234) Vitali Scadden, 2010Vitali Scadden A.D.J. Double-stranded containing multiple IU pairs sufficient suppress induction apoptosis.Nat. Struct. 2010; 1043-1050Crossref (84) Wiatrek 2019Wiatrek D.M. Candela M.E. Sedmík Oppelt Keegan L.P. O’Connell M.A. Activation innate immunity mitochondrial mouse cells lacking p53 protein.RNA. 2019; 25: 713-726Crossref (13) Consistent with this hypothesis, embryonic lethality Adar1 null mice can be rescued concomitant knockout (Liddicoat Scholar) or antiviral signaling (MAVS) (Mannion 2014Mannion N.M. Greenwood S.M. Young Cox Brindle Read Nellåker C. Vesely Ponting C.P. McLaughlin al.The RNA-editing enzyme controls immune responses RNA.Cell Rep. 9: 1482-1494Abstract (331) Scholar), involved response dsRNA. regulation expression also tightly tied (IFN) pathway, gene transcribed isoforms, which, p150, IFN. second isoform, p110, constitutively expressed (Patterson Samuel, 1995Patterson Samuel C.E. Expression double-stranded-RNA-specific deaminase cells: evidence forms deaminase.Mol. Cell. 1995; 15: 5376-5388Crossref (405) pathophysiological, perspectives. Dysregulation common autoimmune disorders (Song 2016Song Sakurai Shiromoto Y. Nishikura Functions Editing Their Relevance Human Diseases.Genes (Basel). 7: 129Crossref (45) viral infections (Pfaller 2018Pfaller C.K. Donohue R.C. Nersisyan Brodsky Cattaneo Extensive cellular accounts proviral activity ADAR1p150.PLoS 16: e2006577Crossref (42) 2011Samuel Adenosine acting (ADARs) proviral.Virology. 2011; 411: 180-193Crossref (201) cancer (Chen 2013Chen Lin C.H. Chan T.H.M. Chow R.K.K. Song Liu Yuan Y.-F. Fu Kong K.L. al.Recoding AZIN1 predisposes hepatocellular carcinoma.Nat. Med. 2013; 19: 209-216Crossref (308) Fumagalli 2015Fumagalli Gacquer Rothé Lefort Libert Brown Kheddoumi N. Shlien Konopka T. Salgado al.Principles Governing Breast Cancer Transcriptome.Cell 13: 277-289Abstract (136) Han 2015Han Diao Yu Xu Werner H.M.J. Eterovic A.K. Genomic Landscape Clinical Cancers.Cancer 28: 515-528Abstract (272) Paz-Yaacov 2015Paz-Yaacov Buchumenski I. Porath Danan-Gotthold Knisbacher B.A. Elevated Activity Is Major Contributor Transcriptomic Diversity Tumors.Cell 267-276Abstract (193) Mutations cause disease Aicardi-Goutières syndrome, (Rice 2012Rice G.I. Kasher P.R. Forte G.M.A. Mannion Szynkiewicz Dickerson J.E. Bhaskar S.S. Zampini Briggs T.A. al.Mutations syndrome associated type I signature.Nat. Genet. 2012; 44: 1243-1248Crossref (505) severe interferonopathy leading neurological damage affected children. linked cancer, its targeting was shown beneficial overcoming programmed cell death 1 (PD-1) checkpoint blockade modulation tumor-specific microRNAs (miRNAs) (Ishizuka 2018Ishizuka J.J. Manguso R.T. Cheruiyot Bi Panda Iracheta-Vellve Miller B.C. Du P.P. Yates K.B. Dubrot al.Loss tumours overcomes resistance blockade.Nature. 565: 43-48Crossref (228) Mehdipour 2020Mehdipour Marhon S.A. Ettayebi Chakravarthy Hosseini Wang de Castro F.A. Loo Yau H. Ishak Abelson al.Epigenetic therapy induces transcription inverted SINEs dependency.Nature. 2020; 588: 169-173Crossref Ramírez-Moya 2020Ramírez-Moya Baker A.R. Slack Santisteban ADAR1-mediated novel oncogenic process thyroid regulates miR-200 activity.Oncogene. 39: 3738-3753Crossref (30) Wei, 2019Wang Wei Deletion ADAR1A: strategy potentiate PD-1 blockade.Signal Transduct. Target. Ther. 4: 6Crossref (3) rendering it attractive target. enzyme, fundamental within dsRNAs what levels remain To date, all addressing question have focused immediate vicinity sites. These revealed degenerate preference, including bias against G position immediately (Eggington 2011Eggington Greene Predicting ADAR RNA.Nat. 2: 319Crossref (214) Matthews 2016Matthews M.M. Thomas Zheng Tran Phelps K.J. Scott A.I. Havel Fisher A.J. Beal P.A. Structures ADAR2 bound reveal base-flipping mechanism selectivity.Nat. 23: 426-433Crossref (97) slight enrichment following (Ouyang 2018Ouyang Z. Ren An Bo Shu W. landscape editome 462 genomes.Sci. 8: 12069Crossref (7) mismatched A-C preferably (Roth 2019Roth quantification adenosine-to-inosine activity.Nat. Methods. 1131-1138Crossref (40) Wong 2001Wong S.K. Sato Lazinski D.W. Substrate ADAR2.RNA. 2001; 846-858Crossref (161) Structurally, known edit secondary (Morse However, some reports indicate few could better than perfect double (Brümmer 2017Brümmer T.W. Xiao Structure-mediated abundance editing.Nat. 2017; 1255Crossref (35) Solomon 2017Solomon O. Di Segni Cesarkas Marcu-Malina V. Mizrahi Stern-Ginossar Kol Farage-Barhom Glick-Saar al.RNA leads context-dependent transcriptome-wide changes structure.Nat. 1440Crossref tertiary may play well (Reenan, 2005Reenan R.A. Molecular determinants guided evolution species-specific editing.Nature. 2005; 434: 409-413Crossref (92) Tian 2011Tian Sachsenmaier Muggenhumer Waldsich Jantsch M.F. Jin determinant required 5669-5681Crossref (25) Attempts characterize effect variations been partially successful individual but not provide comprehensive, globally applicable (Liu 2019Liu Sun Shcherbina Q. Kappel Jarmoskaite Das Kundaje Learning cis-regulatory principles ADAR-based CRISPR-mediated mutagenesis.bioRxiv. : 840884Google Overall, features, even when combined, fail explain predict highly complex landscapes, suggesting involvement more remote and/or elements dictating state levels. Unraveling inosine critical importance perspective. Since genome side effects, has become major recent years, offering potentially safer alternative correct single-nucleotide mutations (Aquino-Jarquin, 2020Aquino-Jarquin Novel Engineered Programmable Systems ADAR-Mediated Editing.Mol. Acids. 1065-1072Abstract (15) Using CRISPR-based (Cox 2017Cox D.B.T. Gootenberg J.S. Abudayyeh O.O. Franklin B. Kellner M.J. Joung CRISPR-Cas13.Science. 358: 1019-1027Crossref (711) modified oligos (Merkle 2019Merkle Merz Reautschnig Blaha Vogel Wettengel Stafforst Precise recruiting ADARs antisense oligonucleotides.Nat. 37: 133-138Crossref complementary (Qu 2019Qu Yi Zhu Cao Zhou al.Programmable using engineered RNAs.Nat. 1059-1069Crossref (70) several groups sought guide either exogenous machinery toward pre-designed targets. endeavors varying degrees were able achieve partial targets basal conditions. improved understanding guiding will help generate modalities abilities. decipher code massively parallel approach, measured variants systematically exploring space 2 reporters. assays permitted discovery principles, independently sequence, editing: deposition orientation-dependent symmetricity. results simple, parsimonious, explaining length structure, allow optimization oligo-based editing. unravel effective develop reporter characteristics: (1) interpretable (2) positions undergoing interdependencies between different positions, least subset high ensure adequate dynamic range. hypothesized long, hairpin satisfy above requirements. chose B2 element, among prime (Neeman 2006Neeman level determined repeat repertoire.RNA. 2006; 12: 1802-1809Crossref (114) designed 146-bp-long stem forming element reverse complement 46-bp loop separating first arms (B2 construct). Ensuring trends observed construct due unique features present particular another identical stem-loop 3? end mNeonGreen (mNG) (mNG negative arm B2/mNG (rather complement) (Figure 1A; Table S1). four plasmids (EF1a promoter) transfected HEK293T cells. mNG contained 14 25 10%, respectively, while no controls. substantial variability strands, being 80%, others 1B). Intriguingly, construct, three clusters (50%–80%) observed, roughly 30- 35-bp them approximately 25, 50–60, 90, areas low (<20%). In addition, appeared deposited strands. This symmetricity apparent only qualitatively quantitatively—in other words, adjacent correlated (R = 0.654, p 3.4 × 10?8) 1C). Interestingly, intrinsic ?35-bp construct. speculated interdependence occur single-molecule level, convoluted hence escape detection, bulk analysis. Given each read stems molecule, analyzed co-occurrence single-read level. Remarkably, unsupervised clustering concordant precisely (A15, 16 A51, 52; A67 A102, 103; A28 A63, 64) 1D). thus motivated dissection observations under perturbational settings. Motivated above, causal formation. Toward end, oligo library disrupted sequences 2A; S2). Each 2,000 replace cloned, pool, plasmid already invariable arm. 8-nt barcode 2B). pools transfected, biological duplicates, untreated, IFN-?-stimulated ADAR1-depleted cells, MCF7 A549 lines NIH 3T3 line. After extraction, transcribed, amplicon libraries prepared sequenced readouts entire 2C). detect 100% samples average depth >3,500 reads per (minimum: 18). showed reproducibility replicates 0.93–0.99, < 2.2 10?16 samples; Figures 2D S1B). Analysis 91 constructs, synthesized barcodes, confirmed depended nearly exclusively itself negligible 0.98–0.99, 2E S1C). IFN stimulation led global levels, 1.65-fold higher compared condition. consistent 2.5-fold total achieved (Figures 2F S1E). medium (1%–50%, average: 1.81-fold increase) conditions, leaving lowly (0%–1%) (>70%) almost unchanged, binding high-affinity (modified conditions) saturating 2G S1D). Conversely, knockdown complete loss demonstrating lines, detected depend 2H). very (Schaffer 2020Schaffer A.A. Kopel Hendel Picardi line catalogue.Nucleic 48: 5849-5858Crossref distribution relative distinct implying defined conserved humans S1F). 41 44 adenosines, respectively. had lowest number As (average 3–4 sites) highest 4–10 2I). Depending 2%–30% molecules zero events, whereas >15 rare 2J S1G). assess requirement series random increasing 10% disruption strandedness reduced mean less half 30% (or higher) abolishment 2K S1H). examined introduction lengths constructs. Unexpectedly, dramatically increased bases minor 3-bp mismatch ?35 +30 3A, 3B, S2A, S3A, S3B). At ?35, median fold change 1.5- 4.9-fold tested +30, ranged 1.4- 2.6-fold lines. absolute terms, difference structurally (? editing) 8.4% 25.7% For instance, 57–59, 24 49%, contrast 7% undisrupted same 85 44% 26% without disruption. parallel, troughs, indicative resistant formed peaks. 0 dropped 5- 7-fold. Furthermore, examination signals peak trough patterns periodic, albeit attenuation magnitude. weak ?70 (in addition ?35), clear peaks intervals, around ?14, ?53, ?87 3B–3E, S2B, Weaker troughs corresponding positive direction; lower magnitude likely consequence editability loop-adjacent largely location disruption, effects Even 1-bp significant 3C S2B). phenomenon pyrimidine-rich bulges (T [data shown], TTC, TTCTT TTCTTCT) 3D 3E). fact offset respect suggestive molecular ruler, form RNA-binding domains (RBDs). periodicity suggests event event. sense, our manner (see Discussion). experiments based single practice, transcriptomes (Barak 2020Barak Finkelstein Roth Purifying defense false immunity.Genome 21: 26Crossref (16) Instead, typically mimic context, devised harboring 1-, 2-, 3-nt selected adenosines. gradually distance mismatches, keeping halfway series, consistently identical, above-described pattern 3F, 3G, S3C). interval remarkably robust tolerates presence intrigued 5-bp most (position ?35) counterpart +30). envisioned possible scenarios underlie observations. model, sense orientation (e.g., loop), determines introduces Under scenario, closer 30-bp offset, beginning 4A, top). senses resp